Hyper-Personalization as biggest AI opportunity

Business and marketing have changed so drastically before our eyes.

I can still remember times when in Poland everything that was displayed in a TV commercial was instantly a successfully sold product. People desired and admired everything they saw on TV. And if it was promoted by a celebrity, it was desired even more.

But those days are gone. TV ads no longer work that well, because we've moved from generalization to personalization. And it's not necessarily because people have changed. It's more caused by new technologies and tools that made it possible.

And the personalization trend has been here for a while now. As soon as social media ads and influencer marketing started, businesses were mentioning personalization, both for B2B and B2C. And this personalization trend is getting more and more important. It's not enough to target "software engineers". You should target "software engineers working with JavaScript for mid-size companies in the US". That's one of the best pieces of advice for enhancing sales, because if a customer feels this service/product is tailored for them, they will be more willing to buy it.

And with all the coding and advanced segmentation tools we can really do a lot in terms of personalization. But AI takes this to a whole new level.

Remember how emails used to be called "personalized" if they started with your name like "Hey Kamil?" and that was all? And this name was sometimes not working when it wasn't a typical name or it was just from other countries? With AI you can easily personalize every email including every detail we have on a customer or based on their previous activities.

We can easily inject personalized content for a website based on who is viewing the site.

It's even possible to create a SaaS that will perform operations differently based on the customer, time of day or any other broader context.

And most importantly - for doing this we no longer need ultra-complex infrastructure with hundreds of predefined scenarios or to write any fancy machine learning ourselves.

The future of hyper-personalization is now.

LLMs (and content vectorization) are the technology we've been dreaming about for so long. Even though software engineers were always great at coming up with and coding different scenarios of customer behavior and validating inputs, working with natural language was always a nightmare.

Especially if we considered different forms and tenses. People make a lot of mistakes. In different languages. With all the typos. Before the LLM era, it was a really horrible experience.

But LLMs made that super easy. And that's why they're perfect for personalization. Now in order to understand a user's specific case you no longer need to consult the CEO, sales representative, 2 developers and a customer service guy. If you have enough data about the user and even their nasty, poorly formatted feedback - AI can understand what happened and how it can be fixed.

And it's no longer acceptable to answer your customer's question with links to hard-to-read documentation pages that even developers hardly understand. AI can help you generate a response in the language the client would understand easily. And allow asking follow-up questions in order to make sure everything is clear.

But personalization can go even beyond that. If we switch from building fully linear software to more agentic ones, where you have a set of pretty universal tools and an LLM trained to understand a specific niche - you can build software that will work differently for different use cases.

How can you enable such hyper-personalization in your app?

Firstly, we need to have some data to work with. So it's a good idea to start collecting:

When having data, it's time to standardize and process the shit out of it. We need to analyze users' behavior and understand where personalization can (and should) be applied and where it shouldn't be.

If you're about to personalize messages to users (emails or in-app messages), you should firstly identify what the personalized message should look like. If you have some kind of template with directions, translate it into a system prompt and adjust until the LLM creates a perfect message without any issues. Test different scenarios to make it almost perfect.

Attention! If the personalized messages are not working correctly every time you test them, don't push them to production. AI is still in its early phase, LLMs can make mistakes. And there's nothing worse than an irritating AI feature. So before sending messages to real users, make sure it's great.

Early phase of technology development is the perfect place for virality. Sometimes one product can go from “undercover” to “hyped” in days or even hours.

Same is with AI. Entire communities are trying different stuff with LLMs, testing products and solutions. And during “creator” times, many people are sharing their thoughts and experiences, for example on X or Reddit. This is a really unique opprotunity for viral marketing.

Usually those were big AI companies who pushed those viral news about their major releases or scandals.

But last days, entire X was flooded with a single, open source tool: Clawdbot.

When I firstly started to spot this tool references I thought it’s a well design viral marketing campaign for a paid tool. And I was shocked that this is all natural virality for a open source tool.

And what’s even funnier, X was flooded with pictures of Mac Mini computers with mention that it was bought for purpose to run Clawdbot. And because of that, Apple was said to experience some mac mini availability shortage (hard to tell if that’s true).

But why this tool became so popular so fast?

Even developers like if something is easy to setup and use. If you can run it on your small computer or raspberry pi within an hour, it’s definitely great developer experience that you want to share with your peers.

It makes it even better if you can access it in a ultra simple way - using your favourite chat apps. No need to create new account, no more UIs to learn.

Easy peasy and worth sharing

There are tens of useful AI assistants. But this one is solving one important problem of other - memory.

You can chat with LLMs via different apps, but it’s rare for AI to keep the long context and memory. Every time you want it to do something, you have to provide full context, guidances, do’s and dont’s. Of couse, when using Claude Code you can build such system with proper set of markdown files, but it requires time.

Clawdbot is your personal assistant straight away. Everything included, keeping the memory.

AI made a big move - it made developers pay for products for coding. But you can’t change your nature. In the world of all paid tools, an open source alternative is like a fresh, cooling wind. You’re definitely more willing to share a free tool rather than a paid one.

If you see everyone using and praising specific tool - you’ll definitely try it out so as not to stay behind.